🤖 Agent: merged by Ludovic's AI agent: verify passed, parity proven (CI or perf-frame-2's local run), before/after numbers and screenshots in the PR body.
Six scenes from one fixed-seed world (seed 8675309): a plains village among
savanna hills, sunset over it and the river, moonrise over its lit square, a
crimson forest with a zombified piglin, tropical fish over a warm-ocean coral
reef, and survival at dusk with a zombie and the full HUD. Plus Horizon's own
title, Singleplayer, Video options and pause screens.
Captured with horizon-ctl screenshot from the offscreen client (fullscreen
2560x1286 on the Mac mini, default graphics settings, GUI scale Auto), scenes
cropped to 16:9 and menus to 16:10 (left-aligned) and scaled to the old sizes
(1600x900, 1280x800). The scene sources are JPEGs in docs/media again, the
names scripts/assets.mjs already reads; the README's WebP copies are the
rebuilt website media. Alt texts and LEGAL.md follow the new pictures.
The launcher screenshots are unchanged: the Mac mini's session is locked, so
no window capture works.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With an explicit Present Mode (Mailbox, Immediate), the VSync option did
nothing: the setting replaced vanilla's choice even with VSync on, so a
window kept running uncapped. Vanilla's VSync means Fifo, in a window and in
fullscreen alike. Now VSync on always presents with Fifo, and the Present
Mode setting chooses only while VSync is off (Auto keeps vanilla's choice:
Immediate, else Mailbox, else Fifo). The choice is one tested function,
window::chosen_present_mode.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The LAN chunk encoder's section cache (#623) and dimension range (#668)
are kept; a cached section's wire bytes use the registry's palette width,
set once per process before any encoder exists, so the cache stays keyed
on the section alone. The sculk scan's palette check takes the u32 state.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Online mode checks accounts since v2026.10.17.07; until #589, players added
before they join get offline UUIDs, with a workaround. Portals follow a
lone player since v2026.10.17.09. Copied game files need v2026.10.17.07 or
newer. Section 10 links the 2026-10-17 benchmark and says plainly that 25
moving players are too many for now. Links to other docs are absolute, so
they work both on GitHub and on the site.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Port CachedUserNameToIdResolver to the dedicated server, in place of the
in-memory name cache: usercache.json in the universe folder is loaded at
startup (bad entries skipped, a bad date warned about), players are added
when they join and kept for a month, and the file keeps the 1000 most
recently used, newest first, as Gson writes it.
The console's op, deop, ban, pardon and whitelist find players by name
through it, in any case. A name it doesn't hold is asked of the profile
service (profiles::find_by_name) on offline servers too, as vanilla does;
one the service doesn't know is the offline player of the lower-case name
on an offline server, and "That player does not exist" on an online one.
tools/test_user_cache_parity.py runs vanilla's server and Horizon's with a
mock profile service and compares the replies, the names looked up and the
files; the dedicated-server harness now compares usercache.json too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Vanilla skips a resource with an invalid path, a data file that does not
parse, one at a time; Horizon's content loader refuses the whole mod
set. carry now keeps them in outside/ with the reason (report
`unreadable`): upper-case paths, JSON that does not parse, shared
tags with "replace": true. A byte-order mark is dropped.
ATM10's 42 carried mods: 19 such files in 7 mods (Mekanism Tools'
ry_UA lang, CTOV's empty tag, Towns and Towers' replacing tags, broken
template pools, recipes and a lang file), and the 42 load as carried.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- perf_bench.py samples the physical footprint on macOS as well as
RSS. RSS counts the pages the allocator freed and marked reusable,
so the six mods' "+1.1 GB" is about 0.5 GB of footprint.
- examples/resource_memory.rs: live heap, peak, RSS and footprint after
each resource loader, for a prepared or a content-mod root.
- BlockRegistry indexes ids by a hash of the state, checked against
the states, instead of keeping every state twice: -22 MB live for
vanilla, -124 MB for ATM10's 42 mods (193,762 states).
- docs/modpacks.md: the per-loader breakdown and the next cut (the
per-state tables parsed whole at startup).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With the wide state ids, ATM10's 42 data-tier mods (193,762 states)
reach the world once these loaders stop refusing them:
- Block sprites: eight texture arrays of 2048 layers instead of four
(9,790 block textures for the 42).
- Per-state tables read as NBT: the table limits cover 2^20 states, and
the collision table in level data uses them.
- Worldgen block tags: 2^20 blocks instead of 8,192 (mineable/pickaxe
holds every stone block of a pack).
- As vanilla's loaders do, a blockstate definition that does not bind,
a missing or broken block model of a mod, and a recipe that does not
parse are logged and skipped instead of dropping every mod (or
stopping the game).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
What works on main today (install, game files, first run, server.toml, the commands, the web panel,
backups, services, Docker built by hand) and, plainly, what doesn't yet: offline mode only, no
published image, no secure chat or locator bar, the open PRs that are coming. The site renders the
guide into /server at build time, so it has one source. The CSP hashes now cover every page's inline
script, so the privacy page's theme script is allowed too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Walk and explore passes at 1 to 100 bots on the ThinkPad (bots on the Mac
mini), vanilla with the locator bar off, plus an explore pass at a rate both
servers keep up with. The README has the method, the tables, the profile of
what remains at 25 walking bots and the earlier runs; docs/server.md section
9 and the Scale row summarise it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The launcher's "Import modpack" needs the pipeline as a library and a way to
get the jars a pack names without redistributing anything.
- The steps move into the library (pipeline.rs) with a line callback, so the
command line and the launcher run the same code; main.rs is a thin CLI.
- fetch: the jars of the portable mods (content, data tier) into
<work>/jars/, the player's own downloads: copied from a folder they have
(by SHA-1, or a CurseForge app instance's minecraftinstance.json), else
from the pack's own link (.mrpack hosts only) or the same file on
Modrinth's public CDN found by SHA-1, kept only when the hash matches.
What only CurseForge has is listed as needing a CurseForge API key.
<work>/fetch.json records each file.
- port: every step after the import for the portable mods (fetch, the
inventory again with the jars read, carry, report).
- A CurseForge zip whose overrides hold Crash Assistant's modlist.json (All
the Mods 10 does) takes its jar names, mod ids and SHA-1s instead of bare
project ids, so Modrinth answers for most files.
- inventory goes on offline when Modrinth cannot be reached.
All the Mods 10 8.3's CurseForge zip: 496 files named; of the 54 data-tier
content mods 42 download from Modrinth's CDN (84 MB) and 12 need the key or
the player's instance; 42 carried over (partial: the harness is a dev step).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Loading the eight carried-over mods in the game (not only the probe) showed
two limits real packs reach. Block textures past 2048 layers are another PR;
this one handles block states and measures a pack.
- Chunk sections store a block state in 16 bits. Vanilla 26.3 has 35,723
states; Macaw's four mods add 36,154, the eight 41,389. Past 65,536 ids
wrapped onto other blocks (a trapdoor became a hanging sign).
mod_content::build now refuses a set of mods past MAX_STATES and names the
states per mod; the game then starts without content mods, as for any
refused content.
- tools/mod_port_harness.py probes each mod on its own against vanilla, as a
player adding it would load it.
- tools/perf_bench.py --game neoforge: Minecraft 1.21.1 on NeoForge 21.1.259
with --mod-jars and --deps, driven by a dev-only NeoForge port of the bench
mod (mods/horizon/src/neoforge-bench); --port-work loads a work folder's
carried-over mods into Horizon. tools/neoforge_oracle.py installs the
client once under .local/neoforge-oracle (the installer, then the vanilla
libraries it inherits, sha1-checked) and builds the driver. A NeoForge
start that crashes before the driver loads (FML reading fml.toml while
creating it) is retried once.
Measured (docs/modpacks.md): six ported mods in Horizon against the same
mods on NeoForge, Mac mini M4 Pro, 2 games per row on a loaded machine. The
world is ready 3.1x sooner at 8 and 16 chunks, the flights run 1.7x and
3.6x the frames, and Horizon's RSS is not lower (+1.1 GB with the mods).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
online-mode=true now authenticates players as vanilla's
ServerLoginPacketListenerImpl does. The server makes a 1024-bit RSA key
pair at startup ("Generating keypair"), sends it with a random challenge,
checks the challenge and decrypts the shared secret from the client's key
packet, encrypts the connection with AES/CFB8, and asks the session
service's verify endpoint (hasJoined, through the services discovery
document) whether the player joined with the server's hash, from the
"User Authenticator #N" thread. A verified profile keeps its UUID and
properties, so skins show in the player list; an unverified one is
refused with "Failed to verify username!", and an unreachable service
with authservers_down. The offline-mode warnings are only printed in
offline mode.
Refused logins are logged as vanilla logs them: "Disconnecting
<name> (/<address>): <reason>" and "<name> (/<address>) lost
connection: <reason>", with log-ips respected.
horizon-protocol gains RSA key generation (Miller-Rabin primes,
e = 65537), the X.509 encoding of the public key and PKCS#1 v1.5
decryption. test_online_mode_parity.py runs its logins against Horizon's
server too, plus a verified player the whitelist refuses, and compares
the two servers' log lines about the player.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`report <work>` writes coverage.json and coverage.md: per mod ported,
partial, not yet or not needed with the reason, and the totals by class,
tier and licence family. It reads the inventory, each mod's carry report
and its harness result: ported is a green harness with nothing left out
(blocks with a class of their own are counted and named); partial lists
the failing checks, the blocks left out and what the templates miss; not
yet names the API tier or the dependency it waits for; not needed covers
libraries, performance mods and pack tooling. Unclassified mods are not
yet, never not needed.
`import` also reads Crash Assistant's modlist.json, the list ATM10
publishes in its repository (mod ids, names, versions, Modrinth hashes,
the loader entry), and the inventory uses its mod ids and names when
there is no jar.
ATM10 (496 files, the eight mods we downloaded carried and checked):
ported 1, partial 7, not yet 367, not needed 121.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The harness the modpack design calls for, for data-tier mods:
- tools/neoforge_oracle.py installs a NeoForge 21.1.259 server (ATM10's)
under .local/neoforge-oracle, builds the exporter in
mods/horizon/src/neoforge-oracle (our code, never shipped), runs it
headless with the user's jars and writes every block (class chain,
states, per-state hardness, light, tool need, map colour, sounds,
outline and collision; drops for hand, diamond tools, shears and silk
touch; placement; BlockSetType and WoodType), item, recipe and tag of
the mods' namespaces, plus the vanilla blocks for templates. About 30 s.
- examples/mod_port_probe.rs builds the content root from the carried
mods and dumps what Horizon's loaders make of the same things.
- tools/mod_port_harness.py compares them per mod and writes
<work>/mods/<id>/harness/result.json. 1.21.1 sound events map to 26.3's
through vanilla blocks both have, and vanilla items newer than 1.21.1
are not counted against a recipe.
On ATM10's own jars (MIT): Macaw's Trapdoors passes everything; Doors,
Fences, Paths, Glassential, Factory Blocks and Cloud Glass pass every
drop, recipe, tag and placement and fail only on what API 0.2 cannot say
(shapes and sound events of their own, blocks of classes no vanilla block
shares). Jars, facts and results stay in the work folder.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Carrying over seven real ATM10 mods (Cloud Glass, Factory Blocks,
Glassential, Macaw's Doors, Fences, Paths and Trapdoors) showed what the
guesses missed:
- Loot tables: 1.21.1's `conditions` lists, `functions` and
`condition`/`function` kinds are now 26.3's single `condition`,
`modifier` and `type`; silk-touch and shears tool matches become
`minecraft:tool/can_silk_touch` and `can_shear`. Before, every drop
condition was ignored (glass dropped itself by hand, upper door halves
dropped doors, double slabs dropped one).
- With the oracle's facts (<work>/oracle/facts.json, from
tools/neoforge_oracle.py in the next PR), every registered block gets
the vanilla template of the same vanilla class and states that matches
its shapes, sounds, BlockSetType and WoodType best, and the mod's own
hardness, blast resistance, light, tool need, friction, map colour and
sounds. Items get stack size and rarity. The carry report lists what a
template does not reproduce and why a block was left out.
- Convention tags a recipe uses (c:dyes/red) are written from the
oracle's tags, vanilla and own members only; the jar's own c tags stay
inside the mod.
- Without the oracle, blockstates that omit properties their models
ignore (powered, waterlogged) still find their template, and names pick
between fences and panes, doors and trapdoors.
- `version = "${file.jarVersion}"` is read from MANIFEST.MF.
- `--resources <prepared>` limits templates to blocks Horizon has.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Loading seven real NeoForge mods carried over by horizon-port-pack (about
1,000 blocks) broke the content root in four places:
- The per-state NBT tables (cached collision, occlusion, movement, worldgen
block properties and support) passed the NBT reader's 1M-node cap once
mod states were appended. They are trusted prepared tables with a row
per state, so they now read and write with Limits::tables() (16M nodes,
256 MiB) everywhere they are loaded.
- The overlay linked the base root by the path it was given, so a relative
base gave dangling links. The base is made absolute first.
- Carried-over tags name blocks the port left out and other mods' items.
Vanilla fails such a tag as a whole; entries naming ids nobody registers
(or tags nobody defines) are now dropped, so the rest loads.
- Mods use the convention tags (c:dyes/red, c:glass_blocks/colorless).
Content mods may add to data/c/tags/ as to vanilla's, and with content
mods loaded the tag loaders read the c namespace and the mods' own.
Stonecutter and smithing recipes in mod namespaces load too (ids
<ns>:<name>), as crafting and cooking ones already did. With no content
mod nothing changes: the loaders read the minecraft namespace only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
What the dedicated server does today (measured against vanilla 26.3), the horizon-server supervisor and
its packaging, server.toml, the CLI and TUI, the web panel, backups and updates, the control API and MCP,
the bot benchmark plan against vanilla and Paper, the working version for 2026-10-23, and the PRs that
build it (#551 to #556).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
carry <work> [<mod id>...] turns each content mod with a jar into a folder
content mod (API 0.2) in <work>/mods/<id>/carry/, for local use:
- assets/<id>/, data/<id>/ and data/minecraft/tags/ are copied; other
namespaces and vanilla data go to outside/ for review; no class file is
copied.
- Vanilla-type recipes get 26.3 ingredients ("id", "#tag", lists);
custom types and NeoForge conditions are kept and listed.
- Item models get the items/ client definitions 1.21.4 introduced.
- Each block is declared from a vanilla template guessed from its
blockstate properties; unknown property sets are listed. Items
without a block are declared from minecraft:stick.
The test builds a small NeoForge jar, carries it over, and reads the
result with horizon-mods' own Content reader.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A CLI for the first steps of docs/modpacks.md's pipeline:
- import: a Modrinth .mrpack, a CurseForge zip or a Prism / MultiMC
instance (or a folder of jars) to pack.json: Minecraft and loader
versions, each file with its hash and sides, and the overrides.
- inventory: Modrinth metadata by SHA-1 (cached in modrinth.json, so
reruns work offline), and the jars when the user's instance has them
(neoforge.mods.toml / mods.toml / fabric.mod.json and a file census,
without decompiling). Each mod gets a class and API tier from the
reviewed ATM10 table, then rules on categories, sides and the census,
and a dependency layer. Writes inventory.json and inventory.csv.
Only metadata is downloaded; jars are never fetched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A .hmod may carry assets/ and data/ beside mod.wasm, or instead of it. The
manifest needs api = "0.2" and the new registries capability ("add blocks
and items to your worlds"). Content mods have no WIT interface: code still
links API 0.1.
horizon_mods::content::Content reads a mod's content and checks it:
- plain resource paths only;
- 65,536 files, 16 MiB each, 512 MiB in all;
- files in the mod's own namespace, plus additions to vanilla tags;
- blocks and items declared in data/<id>/horizon/{block,item}/*.json, each
copying a vanilla template with typed overrides (hardness, resistance,
light, tool, friction, map colour, sounds; stack size, rarity), unknown
keys refused.
Mods::load refuses an enabled mod whose content fails these checks, with
the reason on its row. A content-only mod is on with no engine and no hooks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
docs/modpacks.md: how any pack (ATM10 down to small Fabric packs) gets to
Horizon: import (.mrpack, CurseForge zip or instance, Prism, packwiz),
inventory and classification, dependency order, mechanical carry-over of
data and assets, one agent per mod for code with a per-mod harness against
a dev-only NeoForge/Fabric oracle, and a coverage report. Also the mod API
gaps against 0.1, the registries 0.2 design (template blocks through the
overlay root), licence handling, how "10x" is measured against NeoForge,
and the scope for 2026-10-23.
docs/modpacks/atm10-mods.csv: ATM10's 495 jars from the pack's own
crash_assistant modlist (2026-10-15), with Modrinth metadata looked up by
hash: class, API tier, licence, sides. Metadata only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Nobody could reproduce it, and no issue tracks it, yet every release since
2026-10-14 repeated it. Players noticed (rdbt on Discord).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`wait` ended with release_agent_keys like every other task, so
`key w down` + `wait 40` + `wait 20` walked for the first wait only.
A wait holds nothing: it leaves keys and buttons alone. They stay held
until `key ... up`, `release` or `stop`. Documented in horizon-ctl's
key/wait help (and so the MCP tools) and docs/agent-control.md.
Fixes#469
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Ludovic Marie
LookImEcho-Rick
Cosmic